Higher prices can't crimp server sales as AI drives demand

Global server vendor revenue reached a record $166.3 billion in Q2, up 52% year-on-year, while unit shipments rose 15.4% despite memory shortages and higher component costs. GPU-accelerated servers generated nearly 53% of sector revenue, with average selling prices up almost 44% to $170,200; non-accelerated server pricing rose more than 33% to nearly $13,000. AI infrastructure demand is expanding from hyperscalers to enterprises, neoclouds and government-backed sovereign AI programs, while Dell increased its market share to 13.4% from 7.7% a year earlier as ODM share fell to 53.9%.

Analysis

The investable shift is from a hyperscaler-led component cycle to a systems-integration cycle. DELL is positioned to monetize this better than ODMs and smaller assemblers because enterprise, sovereign and regulated buyers value validated configurations, financing, service and deployment capacity; these attach revenues can support gross-margin resilience even if GPU availability normalizes. The same buyer mix should favor networking, storage and power/cooling content around AI clusters, creating read-through upside for ANET, VRT and ETN rather than a pure server-box trade.

DELL's share gains imply that vendor qualification—not simply aggregate AI demand—is becoming the gating variable. That is unfavorable for white-box exposure and potentially constrains SMCI's premium if customers increasingly prioritize procurement support, long-term servicing and multi-vendor integration; however, SMCI remains the higher-beta beneficiary if accelerator supply improves and lead times compress. HPE has a more mixed setup: sovereign and government deployments fit its sales channel, but its lower apparent share capture suggests execution and AI-system availability must improve before the market awards it comparable multiple expansion.

Over the next 1-3 months, memory pricing is the principal margin risk: systems vendors can pass through component inflation in revenue faster than in gross profit, while customer budgets may ultimately cap unit growth. The 6-18 month upside case is broader inference deployment, which favors lower-cost, non-GPU compute, networking and lifecycle services; the near-term market may be overpaying for accelerator-server ASP while underpricing these adjacent content pools. Falsify the DELL thesis if its next earnings release shows AI-server backlog conversion slowing, Infrastructure Solutions Group margins declining despite revenue growth, or management reduces AI-server shipment/growth expectations.

Key Decisions for Investors

• Initiate/maintain a 3-6 month overweight DELL versus HPE: long DELL / short HPE in equal dollar amounts. The pair isolates superior enterprise AI integration and share-gain exposure; exit if DELL reports weaker AI backlog conversion or HPE demonstrates a material acceleration in AI order growth and segment margin.
• Add a basket of ANET, VRT and ETN on AI-infrastructure pullbacks rather than chase server OEM beta. These names capture recurring network fabric and electrical/thermal content per deployed cluster, with a 6-18 month horizon; key risk is a hyperscaler capex pause or a transition to lower-power inference architectures.
• Avoid adding broad PC-exposed hardware longs until memory costs stabilize. For DELL, require evidence that commercial-client pricing offsets component inflation; otherwise, PC margin pressure can dilute the earnings benefit from Infrastructure Solutions Group growth.
• Use SMCI as a tactical, not core, long only if accelerator lead-time data and quarterly gross margin stabilize. Risk/reward improves on evidence of supply normalization, but a shift toward branded vendors and any further margin erosion would invalidate the trade.